Efficient Lifelong Machine Learning
Eaton, Eric R.
MetadataShow full item record
Lifelong learning is a key characteristic of human intelligence, largely responsible for the variety and complexity of our behavior. This process allows us to rapidly learn new skills by building upon and continually refining our learned knowledge over a lifetime of experience. Incorporating these abilities into machine learning algorithms remains a mostly unsolved problem, but one that is essential for the development of versatile autonomous systems. In this talk, I will present our recent progress in developing algorithms for lifelong machine learning. These algorithms acquire knowledge incrementally over consecutive learning tasks, and then transfer that knowledge to rapidly learn to solve new problems. Our approach is highly efficient, scaling to large numbers of tasks and amounts of data, and provides a variety of theoretical guarantees on performance and convergence. I will show that our lifelong learning system achieves state-of-the-art results in multi-task learning for classification and regression on a variety of domains, including facial expression recognition, landmine detection, and student examination score prediction. I will also describe how lifelong learning can be applied to sequential decision making for robotics, demonstrating accelerated learning for optimal control on several dynamical systems, including an application to quadrotor control. Finally, I will discuss our work toward autonomous cross-domain transfer, enabling knowledge to be automatically transferred between different task domains.
- IRIM Seminar Series